Thanks for your reply. I included the void_cast_register call because the base class is not being serialised by the derived class nor is the derived class being serialised and hence my understanding was that it was needed in this special case. It works with the paramaters to void_cast_register swapped. The include file void_cast.hpp does show the parameters as you state, however the documentation and examples have them in the order (special.html). Can I suggest that this is updated please? Regards, Richard -----Original Message----- From: boost-users-bounces@lists.boost.org [mailto:boost-users-bounces@lists.boost.org] On Behalf Of Robert Ramey Sent: 10 June 2005 16:52 To: boost-users@lists.boost.org Subject: [Boost-users] Re: Boost.Serialization and Pointers to ObjectsofDerived Classes void_cast_register should be necessary only in the most unusual cases. It should happen automatically as part of the "base_object" wrappers call. Also it is used void_cast_register which is the opposite you've of what you've done below. So I would remove it from your example. The export should make up for this. There are number of code paths related to borland 5.64 and borland 5.51. I don't know anything about borland 6. void_cast/export/template instantiation are areas which many compilers have exhibited unexpected behavior. So its quite possible that the code as is can't accomodate BC 6. As far as I know - we've never seen borland 6 or tested it. So if you want to use any part of boost with a new compiler whose ancestor has a history of special requirements, you should be prepared to run the test suite for that compiler. Please do this and post the results. Robert Ramey